Keras-yolo3: I wonder to know the difference between train.py and train_bottleneck.py

[gittigxuy]

Thanks

[stefanbo92]

you can use train_bottleneck if you only want to finetune your model (only train the last layer of the network). The bottleneck version will then precompute values of the second last layer for all your input data. Then, for training only the forward pass through the last layer is needed which will speed up the training time dramatically. For me, training on a CPU only took a couple minutes (also see https://blog.keras.io/building-powerful-image-classification-models-using-very-little-data.html)